Fig. 2: Comparative performance of different models and activity prediction with multifunctionality analysis. | Nature Communications

Fig. 2: Comparative performance of different models and activity prediction with multifunctionality analysis.

From: DLFea4AMPGen de novo design of antimicrobial peptides by integrating features learned from deep learning models

Fig. 2

a The predicted performance on the test set, based on results from 10-fold cross-validation (mean ± s.d. over 10 folds, n = 10). Accuracy, precision, recall, and F1 scores were obtained by the trained BAP-MPB (which are ABP-MPB, AFP-MPB, and AOP-MPB) and the other models32,34, all of which were trained and evaluated on the same dataset. For all four metrics, higher scores indicate better performance. b ROC curves of six models on the ABP, AFP, and AOP datasets, and the area under the curve is AUC. c Predictions made by the ABP-MPB, AFP-MPB, and AOP-MPB models on all 20 datasets, retaining only positively predicted samples. Finally, the intersection of positive predictions from the three models is shown, with the number 4760 at the center representing peptides predicted to be bioactive by all three models. d The number of positive samples in each original bioactive peptide dataset, along with the number of samples that predicted with antibacterial, antifungal, and antioxidant activities (bar chart, left Y-axis), and the proportion of samples with triple activities in each dataset (line chart, right Y-axis).

Back to article page